人脸生成（Face Generation）¶
在该项目中，你将使用生成式对抗网络（Generative Adversarial Nets）来生成新的人脸图像。
获取数据¶
该项目将使用以下数据集：
- MNIST
- CelebA
由于 CelebA 数据集比较复杂，而且这是你第一次使用 GANs。我们想让你先在 MNIST 数据集上测试你的 GANs 模型，以让你更快的评估所建立模型的性能。
如果你在使用 FloydHub, 请将 data_dir 设置为 "/input" 并使用 FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".
In [61]:
data_dir = './data'
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper
helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
#helper.download_extract('celeba', cele_data_dir)
In [2]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')
Out[2]:
CelebA¶
CelebFaces Attributes Dataset (CelebA) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 show_n_images 探索此数据集。
In [3]:
show_n_images = 25
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))
Out[3]:
预处理数据（Preprocess the Data）¶
由于该项目的重点是建立 GANs 模型，我们将为你预处理数据。
经过数据预处理，MNIST 和 CelebA 数据集的值在 28×28 维度图像的 [-0.5, 0.5] 范围内。CelebA 数据集中的图像裁剪了非脸部的图像部分，然后调整到 28x28 维度。
MNIST 数据集中的图像是单通道的黑白图像，CelebA 数据集中的图像是 三通道的 RGB 彩色图像。
建立神经网络（Build the Neural Network）¶
你将通过部署以下函数来建立 GANs 的主要组成部分:
model_inputsdiscriminatorgeneratormodel_lossmodel_opttrain
检查 TensorFlow 版本并获取 GPU 型号¶
检查你是否使用正确的 TensorFlow 版本，并获取 GPU 型号
In [4]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf
# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer. You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))
# Check for a GPU
if not tf.test.gpu_device_name():
warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))
输入（Input）¶
部署 model_inputs 函数以创建用于神经网络的 占位符 (TF Placeholders)。请创建以下占位符：
- 输入图像占位符: 使用
image_width，image_height和image_channels设置为 rank 4。 - 输入 Z 占位符: 设置为 rank 2，并命名为
z_dim。 - 学习速率占位符: 设置为 rank 0。
返回占位符元组的形状为 (tensor of real input images, tensor of z data, learning rate)。
In [5]:
import problem_unittests as tests
def model_inputs(image_width, image_height, image_channels, z_dim):
"""
Create the model inputs
:param image_width: The input image width
:param image_height: The input image height
:param image_channels: The number of image channels
:param z_dim: The dimension of Z
:return: Tuple of (tensor of real input images, tensor of z data, learning rate)
"""
# TODO: Implement Function
image = tf.placeholder(tf.float32, [None, image_width, image_height, image_channels], name = 'real_input')
z = tf.placeholder(tf.float32, [None, z_dim], name = 'z_dim')
lr = tf.placeholder(tf.float32, name = 'learning_rate')
return image, z, lr
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)
辨别器（Discriminator）¶
部署 discriminator 函数创建辨别器神经网络以辨别 images。该函数应能够重复使用神经网络中的各种变量。 在 tf.variable_scope 中使用 "discriminator" 的变量空间名来重复使用该函数中的变量。
该函数应返回形如 (tensor output of the discriminator, tensor logits of the discriminator) 的元组。
In [ ]:
def discriminator(images, reuse=False, alpha = 0.2):
"""
Create the discriminator network
:param image: Tensor of input image(s)
:param reuse: Boolean if the weights should be reused
:return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
"""
# TODO: Implement Function
with tf.variable_scope('discriminator', reuse=reuse):
# Input layer is 28x28x3 or 28x28x1
x1 = tf.layers.conv2d(images, 128, 5, strides=2, padding='SAME')
relu1 = tf.maximum(alpha * x1, x1)
relu1 = tf.nn.dropout(relu1, 0.8)
#print(x1.get_shape())
# 14x14x128
x2 = tf.layers.conv2d(relu1, 256, 5, strides=2, padding='SAME')
bn2 = tf.layers.batch_normalization(x2, training=True)
relu2 = tf.maximum(alpha * bn2, bn2)
relu2 = tf.nn.dropout(relu2, 0.8)
#print(x2.get_shape())
# 7x7x256
x3 = tf.layers.conv2d(relu2, 512, 5, strides=2, padding='SAME')
bn3 = tf.layers.batch_normalization(x3, training=True)
relu3 = tf.maximum(alpha * bn3, bn3)
relu3 = tf.nn.dropout(relu3, 0.8)
# 4x4x512
#print(x3.get_shape())
# Flatten it
flat = tf.reshape(relu3, (-1, 4*4*512))
logits = tf.layers.dense(flat, 1)
output = tf.sigmoid(logits)
return output, logits
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)
生成器（Generator）¶
部署 generator 函数以使用 z 生成图像。该函数应能够重复使用神经网络中的各种变量。
在 tf.variable_scope 中使用 "generator" 的变量空间名来重复使用该函数中的变量。
该函数应返回所生成的 28 x 28 x out_channel_dim 维度图像。
In [91]:
def generator(z, out_channel_dim, is_train=True, alpha = 0.2):
"""
Create the generator network
:param z: Input z
:param out_channel_dim: The number of channels in the output image
:param is_train: Boolean if generator is being used for training
:return: The tensor output of the generator
"""
# TODO: Implement Function
with tf.variable_scope('generator',reuse = not is_train):
x1 = tf.layers.dense(z, 7*7*512)
# Reshape it to start the convolutional stack
x1 = tf.reshape(x1, (-1, 7, 7, 512))
x1 = tf.layers.batch_normalization(x1, training=is_train)
x1 = tf.maximum(alpha * x1, x1)
x1 = tf.nn.dropout(x1, 0.8)
#print(x1.get_shape())
# 7x7x512 now
x2 = tf.layers.conv2d_transpose(x1, 256, 5, strides=2, padding='SAME')
x2 = tf.layers.batch_normalization(x2, training=is_train)
x2 = tf.maximum(alpha * x2, x2)
#print(x2.get_shape())
# 14x14x256 now
x3 = tf.layers.conv2d_transpose(x2, 128, 5, strides=2, padding='SAME')
x3 = tf.layers.batch_normalization(x3, training=is_train)
x3 = tf.maximum(alpha * x3, x3)
#print(x3.get_shape())
# 28x28x128 now
# Output layer
logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=1, padding='SAME')
# 28x28x3 now
output = tf.tanh(logits)
return output
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)
损失函数（Loss）¶
部署 model_loss 函数训练并计算 GANs 的损失。该函数应返回形如 (discriminator loss, generator loss) 的元组。
使用你已实现的函数：
discriminator(images, reuse=False)generator(z, out_channel_dim, is_train=True)
In [16]:
def model_loss(input_real, input_z, out_channel_dim):
"""
Get the loss for the discriminator and generator
:param input_real: Images from the real dataset
:param input_z: Z input
:param out_channel_dim: The number of channels in the output image
:return: A tuple of (discriminator loss, generator loss)
"""
# TODO: Implement Function
g_model = generator(input_z, out_channel_dim)
d_model_real, d_logits_real = discriminator(input_real)
d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)
d_loss_real = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_model_real)))
d_loss_fake = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
g_loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))
d_loss = d_loss_real + d_loss_fake
return g_loss, d_loss
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)
优化（Optimization）¶
部署 model_opt 函数实现对 GANs 的优化。使用 tf.trainable_variables 获取可训练的所有变量。通过变量空间名 discriminator 和 generator 来过滤变量。该函数应返回形如 (discriminator training operation, generator training operation) 的元组。
In [17]:
def model_opt(d_loss, g_loss, learning_rate, beta1):
"""
Get optimization operations
:param d_loss: Discriminator loss Tensor
:param g_loss: Generator loss Tensor
:param learning_rate: Learning Rate Placeholder
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:return: A tuple of (discriminator training operation, generator training operation)
"""
# TODO: Implement Function
t_vars = tf.trainable_variables()
d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
g_vars = [var for var in t_vars if var.name.startswith('generator')]
# Optimize
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
return d_train_opt, g_train_opt
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)
In [18]:
"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np
def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
"""
Show example output for the generator
:param sess: TensorFlow session
:param n_images: Number of Images to display
:param input_z: Input Z Tensor
:param out_channel_dim: The number of channels in the output image
:param image_mode: The mode to use for images ("RGB" or "L")
"""
cmap = None if image_mode == 'RGB' else 'gray'
z_dim = input_z.get_shape().as_list()[-1]
example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])
samples = sess.run(
generator(input_z, out_channel_dim, False),
feed_dict={input_z: example_z})
images_grid = helper.images_square_grid(samples, image_mode)
pyplot.imshow(images_grid, cmap=cmap)
pyplot.show()
训练¶
部署 train 函数以建立并训练 GANs 模型。记得使用以下你已完成的函数：
model_inputs(image_width, image_height, image_channels, z_dim)model_loss(input_real, input_z, out_channel_dim)model_opt(d_loss, g_loss, learning_rate, beta1)
使用 show_generator_output 函数显示 generator 在训练过程中的输出。
注意：在每个批次 (batch) 中运行 show_generator_output 函数会显著增加训练时间与该 notebook 的体积。推荐每 100 批次输出一次 generator 的输出。
In [74]:
def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
"""
Train the GAN
:param epoch_count: Number of epochs
:param batch_size: Batch Size
:param z_dim: Z dimension
:param learning_rate: Learning Rate
:param beta1: The exponential decay rate for the 1st moment in the optimizer
:param get_batches: Function to get batches
:param data_shape: Shape of the data
:param data_image_mode: The image mode to use for images ("RGB" or "L")
"""
# TODO: Build Model
image_width = data_shape[1]
image_height = data_shape[2]
image_channels = data_shape[3]
input_real, input_z, lr = model_inputs(image_width, image_height, image_channels, z_dim)
g_loss, d_loss = model_loss(input_real, input_z, image_channels)
d_train_opt, g_train_opt = model_opt(d_loss, g_loss, lr, beta1)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch_i in range(epoch_count):
step = 0
for batch_images in get_batches(batch_size):
# TODO: Train Model
step += 1
batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim)).astype(np.float32)
# 这里run两次g_train_opt，为了避免辨别器收敛太快，导致生成的图片完全没有能力PK
# 参考 https://github.com/carpedm20/DCGAN-tensorflow
_ = sess.run(d_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
_ = sess.run(g_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
_ = sess.run(g_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
if step % 10 == 0:
train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
train_loss_g = g_loss.eval({input_z: batch_z})
print("Epoch {}/{}...".format(epoch_i+1, epoch_count), "Batch {} ".format(step),
"Discriminator Loss: {:.4f}...".format(train_loss_d),
"Generator Loss: {:.4f}".format(train_loss_g))
if step % 100 == 0:
show_generator_output(sess, 25, input_z, image_channels, data_image_mode)
MNIST¶
在 MNIST 上测试你的 GANs 模型。经过 2 次迭代，GANs 应该能够生成类似手写数字的图像。确保生成器 (generator) 低于辨别器 (discriminator) 的损失，或接近 0。
In [92]:
batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2
mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
mnist_dataset.shape, mnist_dataset.image_mode)
调参记录¶
初始参数：
- batch_size = 64
- z_dim = 100
- learning_rate = 0.001
- beta1 = 0.5
模型架构（filter =5, stride = 1 or 2）：
- Discriminator：28x28x1 ---> 14x14x64 ---> 7x7x256 ---> 7x7x512
- Generator: 7x7x512 ---> 7x7x256 ---> 14x14x64 ---> 28x28x1
第一次结果：Discriminator Loss: 0.8251... Generator Loss: 0.9614 效果不是很好，基本形状出来了，但是图片灰蒙蒙的
调整beta1：
- 0.3 ---> Discriminator Loss: 0.8399... Generator Loss: 0.8489 与0.5的结果接近
- 0.1 ---> Discriminator Loss: 2.1238... Generator Loss: 3.7789 非常糟糕
调整结构：减小辨别器的大小
- Discriminator：28x28x1 ---> 14x14x64 ---> 7x7x128 ---> 7x7x256
- 保持初始参数 ---> Discriminator Loss: 1.6704... Generator Loss: 0.2734
- 辨别器的loss从一开始就比生成器的loss就来得大，不太合理，感觉互换了竞争地位，还是换回之前的模型结构
调整学习率：
- 0.005 ---> Discriminator Loss: 1.1054... Generator Loss: 1.7795
- 0.0005 ---> Discriminator Loss: 0.9000... Generator Loss: 0.6375
- 0.0001 ---> Discriminator Loss: 0.6731... Generator Loss: 0.9542
z_dim：
- 128 ---> Batch 930 Discriminator Loss: 0.1254... Generator Loss: 2.7208 糟糕
- 增加z_dim会更多的增加noise信息
加入dropout感觉提升也不大 Discriminator Loss: 1.1555... Generator Loss: 0.8900
尝试增加迭代次数到3
感觉差别也不大 Discriminator Loss: 0.6617... Generator Loss: 1.0947
CelebA¶
在 CelebA 上运行你的 GANs 模型。在一般的GPU上运行每次迭代大约需要 20 分钟。你可以运行整个迭代，或者当 GANs 开始产生真实人脸图像时停止它。
In [93]:
batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5
"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2
celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
celeba_dataset.shape, celeba_dataset.image_mode)
调参记录¶
更改了模型结构：
- Generator: 7x7x512 ---> 14x14x256 ---> 28x28x128 ---> 28x28x3
- Discriminator：28x28x3 ---> 14x14x64 ---> 7x7x128 ---> 7x7x256
- 学习率 0.00025 ---> Discriminator Loss: 0.4439... Generator Loss: 1.4835 感觉很一般，在2000个batch后，g_loss突然上升了
尝试加入dropout：Discriminator Loss: 1.3863... Generator Loss: 0.4561
最后的模型结构：
- Generator: 7x7x512 ---> 14x14x256 ---> 28x28x128 ---> 28x28x3
- Discriminator：28x28x3 ---> 14x14x64 ---> 7x7x128 ---> 4x4x256
为了查看下模型的最终效果，将迭代次数调整到2次，效果不错，基本看到了人的五官
迭代次数继续增加，应该会得到更好的结果
提交项目¶
提交本项目前，确保运行所有 cells 后保存该文件。
保存该文件为 "dlnd_face_generation.ipynb"， 并另存为 HTML 格式 "File" -> "Download as"。提交项目时请附带 "helper.py" 和 "problem_unittests.py" 文件。